When you’re planning a neighborhood, you start with big ideas and blank spaces. Should it be tall and dense or low and leafy? Where do the shops go? How much green space is enough? All those choices shape how people will live, and how much energy they will use. It sounds a little like the game SimCity, but in the real world, the environmental consequences of those choices are harder to see.
Simi Hoque, PhD, professor of civil, architectural and environmental engineering at Drexel University, is changing that. Her research team has developed a machine learning tool that predicts neighborhood-scale energy use and carbon impact using only a handful of basic inputs. The goal is to give planners useful feedback before the real details, like shape, size and final use, are even on the table.
“You can start with something really simple,” Hoque said. “Like, ‘I want this development to be 20 percent office, 30 percent residential, 50 percent commercial, in this region.’ Press go. And it gives you a rough estimate of what the carbon footprint would be.”
The tool is built for the sketch stage. You can start with simple inputs, like the mix of homes, offices, and shops and the location, then let it show where trouble might crop up. It may warn that allowing much more floor area for housing in a given spot could push home energy use higher. It may flag a plan that leans on big office or retail spaces as a likely driver of higher total energy. Move the same idea to a colder climate zone and it will show a jump in heating demand. You can nudge the inputs and see the signals change in minutes, which makes these readouts useful for direction even if they are not the final engineering numbers.
It can generate results from basic ratios and location data, then refine those results as more details come in. By compressing what used to take weeks into a matter of minutes, it allows more room for experimentation.
“A lot of this type of work usually takes a very long time, with complicated math that not everyone has access to,” Hoque said. “We wanted something more dynamic. Something that could keep churning out new solutions and show what the implications are, without getting stuck on the math.”
The model was trained on publicly available datasets from the U.S. Department of Energy and the Census, drawing from thousands of real buildings to learn patterns in land use, population density and building activity. It takes in high-level variables such as floor area by building type, climate zone and development density, and outputs estimated site energy use and greenhouse gas emissions.
Hoque’s team validated the model using a test set of hypothetical neighborhood designs created in Rhino, a popular digital modeling platform. Each scenario varied in form and mix — some more residential, others more commercial — allowing the tool to demonstrate how design choices shift projected energy profiles.
“At this stage, the tool is not about architectural detail,” Hoque said. “It’s about understanding the broader configuration, the type and intensity of use, the climate zone, and giving you a sense of the emissions profile before you lock in those big decisions.”
The concept began during Hoque’s Fulbright residency at Politecnico di Milano, where she worked alongside researchers focused on neighborhood-scale decarbonization. That experience helped shape the vision for a lightweight planning tool that could integrate with common design workflows rather than replace them.
Back at Drexel, her team adapted the system to match U.S. planning norms. The model now works with any of the country’s sixteen climate zones and can produce meaningful estimates even when major design details are still in flux.
“We’re hoping it helps cities or developers that are maybe a little more forward-thinking,” Hoque said. “Who want to design neighborhoods that are more climate-ready, more efficient, lower in greenhouse gas emissions.”
The tool is already drawing outside interest. One of the firms in discussion is Arup, a global design and engineering consultancy. Other potential collaborators include local planning agencies and sustainability offices looking for more agile forecasting tools.
The model is also designed to be transparent. Users can see which variables are driving its predictions, allowing for clearer comparisons across design scenarios.
“We made a conscious decision to focus on interpretability,” Hoque said. “Because we want the user to understand why the model is giving them that number, not just take it on faith.”
She believes that kind of visibility is missing in most planning processes today, particularly in the United States. In parts of Europe, new developments must already include projections of energy use and carbon output as part of regulatory review. The United States is far behind on that front.
“In Europe, they have to analyze whether the carbon footprint of a new design meets certain thresholds,” she said. “They have to stipulate how much energy it will take to power the complex.”
“In the U.S., I don’t know of any cities that require that.”
Rather than wait for regulation to catch up, Hoque hopes to put better tools in the hands of designers and planners now.
“It’s kind of like giving them a calculator,” she said. “They still have to do the work, but now they have something that makes the feedback faster and more useful.”
Her lab is already planning the next two phases of the system, which will allow for more detailed modeling of building geometry and block-scale energy systems. But even in its current form, the tool provides something new: a way to test big ideas, quickly and with confidence.
“We just want to make it easier for people to make better choices,” Hoque said.
